Manufacturing Intelligence Beyond the Factory: Supplier Quality and Risk

By Harri Pihlajavesi, CEO | Predisys

A large part of manufacturing quality depends on processes you don't directly control.

Components, materials and subassemblies arrive from suppliers, sometimes through several tiers of a supply network. By the time a supplier problem becomes obvious, the affected material may already be in production, incorporated into finished products or moving toward customers.

I saw an interesting example of this in electronics manufacturing involving printed circuit board assemblies.

Automated test data began showing a statistical change in the behavior of a component. The products were being tested as part of normal manufacturing, and something in the results had started behaving differently from what we had seen before.

The change was investigated and eventually led to the discovery of counterfeit components in the supply chain.

Nobody looked at the original statistical signal and concluded that the components were counterfeit. The signal gave us evidence that something had changed. From there, people could begin working out what it was.

I think that distinction is particularly useful when we talk about supplier quality because the first evidence of a supplier problem may appear in our own manufacturing processes long before we know what caused it.

A supplier change can appear in your process

We usually have a considerable amount of information specifically associated with suppliers. There are incoming inspections, certificates, audits, nonconformances, corrective actions, supplier ratings and other quality records.

Once a component or material enters production, we begin collecting another kind of information about it.

We see how the process behaves while using it.

The effects may appear in dimensional measurements, process parameters, automated test results, yield, scrap, rework or eventually in product performance.

Suppose a new supplier lot enters production. The material passed incoming inspection and meets its documented requirements, but several days later a test parameter begins to shift.

The products may still be passing. Overall yield may still look normal. There may be no customer complaint and nothing that would normally be classified as a supplier quality problem.

If the test process has been stable for a long time, however, the change in its statistical behavior may be worth investigating.

We would want to know when the change began and what was happening in production around that time. Did a new material or component lot enter the process? Was the behavior limited to products containing that lot? Did anything else change on the line? Are other characteristics moving with it?

The supplier information becomes part of the manufacturing context we use to answer those questions.

Following the material through manufacturing

Traceability becomes especially useful here.

If a supplier lot is suspected, we want to know where it went. Which production lots used it? Which products contain it? Which lines or plants received it? Is there still material in inventory?

Then we can look at what happened in those places.

Perhaps every product containing the material behaved normally except one. Maybe the statistical change appears across several products using the same component. It may be present at one plant and absent at another.

Traceability tells us where the material went, and the manufacturing history tells us what happened when it got there.

The relationship works in the other direction as well.

We may begin with a process or test result that looks unusual and work backward. Which products are affected? What do they have in common? Did they use the same component, material lot or supplier?

An investigation can move through these relationships in either direction depending on where the first evidence appears.

Statistical changes can give us time

The timing can make a considerable difference in a supplier problem.

A process can begin behaving differently while the product is still within specification. A test distribution can shift before individual units begin failing often enough to affect overall yield.

If we recognize the change at that point, the affected material may still be in production or inventory and the number of products involved may be relatively small.

Later, the same problem can involve finished inventory, sorting, rework, production interruptions, expedited replacement material, shipments, warranty claims or customers.

This is one of the places where Cost of Poor Quality can grow very quickly.

It is also why understanding normal process behavior is so valuable. If we have a statistical history for a process or test, we have something to compare current behavior against. We don't have to wait for a large number of failures before asking why the process changed.

That doesn't mean every statistical change is a supplier problem. Most won't be. It gives the people investigating the process another piece of evidence and, sometimes, an earlier one.

The pattern may be larger than one product

Supplier quality becomes more complicated when the same material or component is used in many places.

Imagine that a component is used in ten products across three plants.

One product begins showing a small change in a test parameter. At another plant, rework increases slightly. Somewhere else, there are occasional failures that don't yet look like a pattern.

Each team may investigate its own process independently and come to reasonable conclusions based on what it can see locally.

The common supplier relationship may be much harder to see.

Once we know that one process is behaving unusually, we can look for other products that use the same component or supplier lot and ask whether anything similar is happening there.

Sometimes the answer will be no, which is useful information too. It helps narrow the investigation.

Other times, several small changes that looked unrelated when viewed separately may turn out to share something in common.

For a supplier quality engineer responsible for a large network, being able to investigate across those relationships can be much more useful than waiting for every plant or product team to encounter a clear failure on its own.

What the investigation teaches us

Once a supplier problem has been understood, there is another question worth considering: what happens to everything we learned?

Perhaps the supplier changed a manufacturing process without communicating it. A tooling problem affected certain lots. A material property changed. A component came from an unexpected source. In the electronics example, the investigation eventually found counterfeit components.

There will usually be quite a bit of information surrounding the conclusion.

We know which statistical behavior first attracted attention. We know which products and lots were affected, what people investigated, what evidence they found and what action was taken.

Then manufacturing continues and gives us another piece of information: what happened afterward.

If the process returns to its previous behavior after the suspect material is removed, that becomes part of the history. If the problem continues, the investigation needs to go further.

Months later, a similar statistical pattern may appear again.

At that point, the earlier supplier investigation could be extremely useful, provided we can find it and understand why it relates to what we are seeing now.

This is how supplier investigations become part of Manufacturing Intelligence. The organization retains the connection between what happened in the process, the material and supplier involved, what people discovered and what happened after they acted.

Looking across a supplier network

The same information becomes interesting at a much larger scale.

A global manufacturer may have hundreds or thousands of suppliers feeding many products and factories. Supplier quality teams already use audits, scorecards, incoming quality, nonconformances and other measures to understand supplier performance.

Manufacturing behavior can add another view.

A supplier quality engineer could look for unusual statistical behavior associated with components from a particular supplier. If something appears, the investigation could move into the products, plants and supplier lots involved.

The engineer might want to know whether the same pattern is appearing anywhere else, whether it began after a particular lot entered production or whether similar behavior has previously resulted in a confirmed supplier issue.

The significance will depend on the context. A small statistical change involving a critical component used across several high-volume products may deserve more attention than a larger change isolated to a low-volume process.

This is the kind of prioritization supplier quality teams already do. Connecting supplier information with manufacturing behavior gives them more information to do it with.

Following the investigation with AI

These investigations can require a lot of navigation across systems and data.

An engineer may start with an unusual test result, determine which production lots are affected, identify the components those lots have in common, trace the components to supplier lots and then find every other product where those lots were used.

After that, someone still has to determine whether those other products are showing similar behavior.

An AI agent working with Manufacturing Intelligence could help follow those relationships.

A supplier quality engineer might begin by asking:

Are we seeing unusual manufacturing behavior associated with this supplier?

If something appears, the next question might be which supplier lots are involved and where else those lots were used. From there, the engineer may want to know whether the same statistical pattern is appearing in those products or whether the organization has investigated something similar before.

The questions develop as the investigation develops.

An AI agent can help bring the relevant manufacturing, statistical, traceability and supplier information together, while the people investigating still need to evaluate what the evidence means.

A shared supplier does not establish that the supplier caused a process change, just as two things changing at approximately the same time does not establish causation. The relationships tell us where it may be useful to look.

That alone can save a considerable amount of time when the alternative is searching across products, plants, lots and systems manually.

Finding the change earlier

What stays with me about the counterfeit-component example is the point at which the problem first became visible.

It wasn't a customer complaint or a large increase in failed products. The manufacturing data began behaving differently.

There was enough statistical history to recognize the change, and enough context around the data to investigate where it came from.

Supplier quality problems will continue to be found through incoming inspection, audits, failures and all of the other methods manufacturers already use. Manufacturing history gives us another place to look, including changes that appear while products are still being made and still passing.

When process behavior, traceability, supplier information and previous investigations remain connected, we can follow those early signals much further.

For a supplier quality organization responsible for hundreds of suppliers and thousands of components, that can mean finding a problem while the manufacturing process is beginning to tell us about it, rather than learning about it later from everything the problem caused.